Bottom-up ontology work starts with semantic lifting

The opening of Chapter 7 draws a line worth keeping: bottom-up ontology development is not schema export. It is semantic lifting.

Hadto wants to turn messy operating knowledge into reusable business infrastructure. If the platform mistakes today’s software shape for the structure of the business itself, it will formalize storage habits instead of domain meaning.

Why export is not enough

Real business systems are full of implementation artifacts: denormalized tables, legacy compromises, overloaded status fields, foreign keys that hint at relationships, opaque identifiers, and partial mappings from imported formats.

A direct conversion can turn all of that into something formal-looking while still missing the point. Before a source becomes useful ontology input, someone has to recover the concepts, relations, constraints, and vocabulary decisions the implementation only expresses imperfectly.

The first questions are practical: which stored structures reflect domain entities rather than technical convenience, which values are really controlled vocabularies or taxonomies, which rows should remain instance data, which mappings are faithful and which are approximate, and which source assumptions need to stay visible after conversion.

Without that step, the ontology inherits the source system’s accidents.

Why Hadto should treat this as governance

Hadto’s platform needs to learn from databases, documents, imported ontologies, and other imperfect artifacts without letting them dictate the conceptual model.

If semantic lifting is skipped:

• reusable operator infrastructure is built on implementation noise,
• domain experts struggle to recognize their business in the model,
• apprentices inherit concepts that were never made explicit,
• later automation has to reason over yesterday’s storage shortcuts.

Three operating rules

1. Preserve provenance

If an ontology came from a relational schema, an OBO import, or another source representation, the platform should keep that visible. A generated OWL artifact can look final even when it is only one projection of the source. Reviewable systems should record the source, the transformation path, and any known semantic compromises.

2. Make class-vs-individual decisions explicit

Bottom-up ingestion often exposes values that sit between ordinary records and hidden category systems. A source table might have one status column with values like lead, quoted, won, and lost. Those values are useful in the application, but they should not automatically become ontology classes just because they recur in many rows. They may describe stages in a sales process, they may collapse several different business decisions into one field, or they may reflect a workflow the company already wants to replace. Semantic lifting means reviewing that field before promotion: decide whether it represents a controlled vocabulary worth modeling, whether some values should stay instance data, or whether the whole column is just an application shortcut. If the platform does not record when values become ontology classes, remain individuals, or stay application-layer enumerations, the choices still get made. They just become invisible.

3. Keep review surfaces label-aware

A converted ontology may be semantically acceptable and still be unreadable if every interface shows raw identifiers instead of meaningful labels. Hadto needs domain experts, operators, and apprentices to inspect imported assets without already knowing namespace conventions.

What Hadto should preserve

The job is not to export whatever the source system already says. The job is to recover the business meaning inside it.

That is the operating standard: recover concepts before formalizing them, record mapping and conversion decisions, preserve provenance, make class-vs-individual choices reviewable, and present imported knowledge in a form humans can read. If a source system cannot survive that review, it is not ready to shape the ontology.